Deprecated Gen.small_int changed distribution

[jmid]

I just noticed something about the deprecated Gen.small_int...

First of all:

• It is aliased to Gen.small_nat
• It is marked with a note concerning backwards compatibility:
  

  qcheck/src/core/QCheck.ml

  Lines 184 to 186 in b065a81

  	(* NOTE: we keep this alias to not break code that uses [small_int]
  	for sizes of strings, arrays, etc. *)
  	let small_int = small_nat

Unfortunately Gen.small_nat changed distribution from the interval [0;10000[ to the interval [0;100[ with QCheck version 0.10. Since version 0.5.2 Gen.small_int and Gen.small_nat's distributions were caused by a chain of aliases: Gen.small_int = Gen.small_nat = Gen.nat:

qcheck/src/core/QCheck.ml

Lines 97 to 105 in 52e4193

	(* natural number generator *)
	let nat st =
	let p = RS.float st 1. in
	if p < 0.5 then RS.int st 10
	else if p < 0.75 then RS.int st 100
	else if p < 0.95 then RS.int st 1_000
	else RS.int st 10_000
	let small_nat = nat

QCheck version 0.10, introduced the Gen.small_nat / Gen.nat / Gen.big_nat distinction and broke the last alias, thus

• changing Gen.small_nats distribution explicitly
• changing Gen.small_ints distribution as a side-effect:
  

  qcheck/src/core/QCheck.ml

  Lines 97 to 99 in a3c8618

  	let small_nat st =
  	let p = RS.float st 1. in
  	if p < 0.75 then RS.int st 10 else RS.int st 100

A bit of digging tells me Gen.small_int had the distribution in the interval [0;10000[ since version 0.5, it was briefly changed in 0.5.1 to output negative ints (#10), and then restored in 0.5.2 which also introduced the deprecation.

I found out, because rerunning some of my pre-0.10 code using Gen.small_int didn't find the expected counterexamples anymore.

With QCheck tests depending on distributions, a narrower distribution imply that test inputs are confined to a smaller subset.
Potentially this could mean bugs flying under the radar...
If someone is indeed using Gen.small_int for sizes of strings, lists, etc. as the compatibility note says, an upgrade from pre-0.10 means they silently started confining their tests to, e.g., strings of size [0;100[ compared to the earlier [0;10000[.

I see a couple of possibilities from here:

• do nothing - after all, this change in distribution was introduced with 0.10 in 2019, keeping this issue for documentation
• make Gen.small_int an alias to Gen.nat: let small_int = nat
  thus restoring the old distribution in the interest of backwards compatibility
• remove Gen.small_int: it is unclear whether anyone else noticed. Also: it has been deprecated since January 2017.
  Removing it would force users to notice and act accordingly.